O2O Orders' Cooking Time Prediction Based On Text Mining And Ensemble Learning

Rencently,the rapid development of 020 leads to the explosive growth of 020 on-demand delivery.At present,most 020 platforms basically use order dispatch mode.In order to make order dispatch system more reasonable and efficient,the platform needs to predict the 020 orders' cooking time accurately.The accuracy of the orders' predicted cooking time directly determines the efficiency of the dispatch system and plays a vital role in the transport capacity allocation.Firstly,the theory of text mining and ensemble learning is expounded.Secondly,taking internship program as a case,this paper gives the definition of the 020 orders' cooking time and proposes the data cleaning method.After that,this paper builds basic features based on business experience,and uses text mining technology to construct text features as well as uses Early Fusion and Late Fusion to do feature fusion.Then,the paper compares the prediction accuracy of four basic models?of XGBoost,Random Forest and GBRT,and confirms the final fusion scheme of four basic models.Finally,based on MSE and other indicators,the accuracy of fusion model is evaluated,and an online real-time prediction scheme is given.The research results in this paper show that after introducing the randomness,the weighted fusion of multiple ensemble models has better prediction effect than single ensemble model.At the same time,it is proved that the weighted fusion of the model has higher performance than Stacking when the data volume is large,the feature dimension is high,and the computing resources are relatively limited.The final model improves the efficiency of the entire order distribution system,as well as improves the satisfaction of riders and the customers.